Unveiling the geographic distribution of Boana pugnax ( Schmidt , 1857 ) ( Anura , Hylidae ) in Venezuela : new state records , range extension , and potential distribution

Boana pugnax is a treefrog inhabiting open lowlands from southern Central America and northwestern South America. Its geographic distribution in Venezuela is poorly understood due, in part, its morphological similarity with B. xerophylla (with which is frequently confused) and the few localities documented. In order to increase the knowledge of the distribution of B. pugnax in the country, we examined the specimens of B. pugnax and B. xerophylla deposited in 4 Venezuelan museums, compiled the locality records of B. pugnax, and generated a model of potential distribution for species. We report 46 new localities for the species in Venezuela, including 8 new state records, which increases considerably its range extension.


Introduction
Boana pugnax (Schmidt, 1857) is a nocturnal, medium-sized treefrog, inhabiting in open lowlands of southern Central America and northern South America (Frost 2017).Its type locality is "along the Chiriquí rivers not far from Bocas del Toro" (Panama) according Kluge (1979).This species has a discontinuous pattern of geographic distribution, ranging from central Panama, with a hiatus in eastern Panama and western Colombia, and continuing through northwestern Colombia to northwestern Venezuela (Kluge 1979, La Marca 1996, Mijares-Urrutia and Arends 1999, Duellman 2001, Lynch and Suárez-Mayorga 2001, La Marca et al. 2010), between 5 and 540 m above sea level (a.s.l.) (Lynch and Suárez-Mayorga 2001).
Boana pugnax was resurrected from the synonymy of B. crepitans (Wied-Neuwied, 1824) by Kluge (1979), and both species were proposed as members of B. faber group (Faivovich et al. 2005).Recently, Orrico et al. (2017) NOTES ON GEOGRAPHIC DISTRIBUTION resurrected B. xerophylla (Duméril & Bibron, 1841) from the synonymy with B. crepitans.They defined as Boana xerophylla all the populations previously referred as B. crepitans from eastern Panama, Colombia, Venezuela, the Guianas, and northwestern Brazil.Boana pugnax and B. xerophylla are found in sympatry in some localities in Central America and northern South America (Kluge 1979, Lynch andSuárez-Mayorga 2001), and they are frequently confused due its morphological similarities (La Marca 1996), particularly when are preserved.Consequently, numerous specimens of B. pugnax deposited in Venezuelan museums have been previously misidentified, and the current knowledge about its geographic distribution is limited.
In Venezuela, B. pugnax was recorded for first time by La Marca (1996), based on a single specimen (ULABG 3292) from south of Maracaibo Lake Basin (Mérida state).Subsequently, Mijares-Urrutia and Arends (1999) documented a new locality for this species in Falcón state, northern Venezuela, and Chacón-Ortiz et al. (2004) referred it from Táchira state in the eastern piedmont of the Andes (Orinoco basin).The species also was referred by Barrio-Amorós (2004) for the Venezuelan Llanos, but neither precise locality nor voucher specimens were mentioned.Tárano (2010) described the vocalizations of B. pugnax from Guárico state in the Central Llanos from Venezuela, and recently Infante-Rivero and Velozo (2015) recorded this species for the Zulia state, at west Lake Maracaibo.Here we aimed to increase the knowledge about the geographic distribution of B. pugnax in Venezuela.After a review of 4 Venezuelan museums, we report new state records and distribution extensions for B. pugnax in the country.Additionally, we developed a species distribution model, in order to increase the knowledge of the distribution of this species, as well as identify gaps in the potential distribution.

Methods
Venezuelan records of B. pugnax were obtained from the following Venezuelan museums: Colección de Anfibios del Laboratorio de Biogeografía de la Universidad de Los Andes (ULABG), Mérida; Colección de Vertebrados de la Universidad de Los Andes (CVULA), Mérida; Museo de Ciencias Naturales de Guanare (MCNG), Guanare; and Museo de Historia Natural La Salle (MHNLS), Caracas.We verified the identity of all specimens of MHNLS identified as B. pugnax, B. xerophylla (previously B. crepitans,) and Boana sp., and all specimens identified as B. xerophylla and B. pugnax in the other 3 collections, in order to detect specimens of B. pugnax yet unidentified or misidentified, and to recover as many records as possible.
We followed the morphological definition and diagnosis established by Kluge (1979), andDuellman (2001) to identify the specimens as B. pugnax.According these authors and Orrico et al. (2017), adult males of B. pugnax are larger than B. xerophylla [68.9-77.9mm vs 42.9-63.8mm of snout-to-vent length (SVL), respec-tively].Further, the former species has less pigmentation in the anterior half of palpebral membrane, has darker and thicker bars on the pleural region, and dark bars on the anteroventral surface of thighs (Fig. 2).We also considered the differences in throat pigmentation noted by La Marca (1996) between B. xerophylla and B. pugnax from western Venezuela: unpigmented in B. xerophylla and blackish-brown colored in B. pugnax (Fig. 2).These morphological definitions are congruent with molecular (DNA sequences) and acoustic (advertisement calls) evidences obtained previously with specimens from western Venezuela (M.Escalona et al., unpublished data).
We generated the species distribution model (SDM) for B. pugnax based on all the locality records from Venezuela obtained from the museums listed above and from literature (La Marca 1996, Mijares-Urrutia and Arends 1999, Chacón-Ortiz et al. 2004, Tárano 2010), whereas those from Panama and Colombia were compiled from the Círculo Herpetológico de Panama's webpage (Círculo Herpetológico de Panamá 2017), Batrachia online database (Acosta-Galvis 2017), and literature (Kluge 1979, Lynch and Suárez-Mayorga 2001, Armesto et al. 2009, Mendez-Narvaez et al. 2014, Guarnizo et al. 2015).We did not include the records of GBIF due to difficulty of verifying the identity of specimens on which those records are based and we presumed the occurrence of an important number of misidentifications (see below).Each locality was verified and georeferenced (when necessary) using Google Earth ® .All geographic coordinates were transformed in decimal degrees, based on the WGS 84 datum.
We performed the SDM under the Maximum Entropy algorithm in MaxEnt version 3.3.3k(Phillips et al. 2006) due to its better proved performance, including a lower commission error (i.e., overestimating) in the models (e.g., Elith et al. 2006, 2011, Peterson et al. 2007).We used interpolated climate data from WorldClim project (Hijmans et al. 2005) and the terrestrial ecoregions (Olson et al. 2001), both at 30″ resolution (~1 km 2 cell size), as input variables in order to characterize the ecological niche and potential distribution of the focal species.Bioclimatic layers were cropped to include latitudes 01.1947° N to 12.4193° N and longitudes 089.8861°W to 059.9454° W. Despite the fact that categorical variables, such as ecoregions or ecosystems are not commonly used in SDM studies, we decided to include it because this predictor may be considered as variable related directly with the ecology of the focal species (Pearson and Dawson 2003, Austin 2007, Rödder et al. 2009, Jiménez-Valverde et al. 2011).Thus, the obtained models provide a first approximation indicating the better hypotheses for the potential distribution of B. pugnax at the current time.
We used a geographical clip (Fig. 1) based on the intersection of Terrestrial Ecoregions (Olson et al. 2001) and the Biogeographical Provinces of the Neotropic (Morrone 2014) Barve et al. 2011).In addition, for a first explorative analysis, we used the 20 bioclimate layers and assessed which variables were the most important for the model, according to the Jackknife test calculated in MaxEnt (Berger 2007, Elith et al. 2011).In a second modelling exercise, we generated the species distribution using non-correlated environmental variables (r < 0.8) in combination with the most relevant environmental variables identified in the first approach (e.g., Ortega-Andrade et al. 2015).These additional steps allowed us to reduce the collinearity of variables and over-fitting of the generated distribution models (see Peterson et al. 2011).
The SDM was generated with 80% of the locality records (training data) while the other 20% was used for model evaluation (testing data).We ran 1,000 iterations, with no extrapolation in order to avoid artificial projections of the extreme values of the ecological variables (Elith et al. 2011, Stohlgren et al. 2011).All other parameters in MaxEnt were maintained as default settings.We ran 10 cross-validate replicates to calculate confidence intervals in order to select the best model based on the performance of area under the curve or "AUC" (Elith et al. 2006(Elith et al. , 2011)).
Then, we used the logistic response to obtain the values for habitat suitability (continuous probability from 0 to 1; Phillips et al. 2006), which were subsequently converted to binary presence-absence values on the basis of the established threshold value, defined herein as the "Fixed Omission Value 5" or FOV5 (Liu et al. 2013).This threshold allowed us to evaluate the species' distribution by minimizing commission errors (i.e., over-predictions) in our final binary maps.Finally, the performance of the selected model was assessed based the commission and omission errors (Anderson et al. 2003) and by applying the Partial-ROC (Receiver Operating Characteristic) curves test (Lobo et al. 2008, Peterson et al. 2008).This criterion was used to solve problems associated with an inappropriate weighting of the omission and commission errors during the AUC analysis (see Lobo et al. 2008, Peterson et al. 2008).

Results
Our study includes new information about the B. pugnax's distribution, with 179 unique occurrences recorded (57 points from Venezuela, 117 from Colombia, and 5 from Panama), including localities quite outside from the recognized geographic distribution range (Fig. 1, Table 1, Table A1).Surprisingly, we found an important number of misidentifications in the Venezuelan museums reviewed (ca 79%).A total of 46 new localities were recorded for first time for Venezuela, extending the geographic distribution of B. pugnax ca 852 km east of the closest know locality in the country (Fig. 1, Table 1).

Discussion
The new occurrences, based on confirmed identification of specimens, include new state records for Apure, Anz oátegui, Barinas, Cojedes, Guárico, Monagas, Portuguesa, and Trujillo.Thus, our new records change considerably the previous statement of B. pugnax as restricted to western Venezuela (Duellman 2001, Lynch and Suárez-Mayorga 2001, Chacón-Ortiz et al. 2004, La Marca et al. 2010), and exceed the prediction formulated by La Marca et al. (2010), who suggested that this species was expected to be in Apure, Barinas, and Zulia state.
All the 116 specimens recorded (representing 57 unique localities) for B. pugnax in Venezuela are between 0-605 m a.s.l.(Table 1, Appendix Table A1).These records reduce scarcely the elevational range reported previously for Venezuela (0-700 m a.s.l.) by Barrio-Amorós (2004).Other localities documented in Panama are below 100 m a.s.l.(Duellman 2001), while in Colombia are mainly below 500 m a.s.l.(Lynch and Suárez-Mayorga 2001).Thus, this species seems be predominantly from lowlands as suggested Kluge (1979) and Lynch and Suárez-Mayorga (2001).In this sense, it is important to note that our model did not predict 4 localities reported with higher elevation in Colombia (see Table A1 (Mendez-Narvaez et al. 2014).Despite that we argue the need of validation at field for model in these areas, an alternative explanation for these results may involve the misidentifications for the specimens as we found in the Venezuelan museums (see above).
Interestingly, the most important variable found here (i.e.mean temperature of coldest quarter) support the Lynch and Suárez-Mayorga (2001) affirmation that B. pugnax may be "... unable to breed in cooler environ-ments…".The B. pugnax distribution range (according the FOV5) totalled ~658,400 km 2 along the region from Venezuela to Panama, with ~394,300 km 2 (59.9%) estimated for Venezuela, ~245,400 km 2 (37.3%) for Colombia, and only ~18,700 km 2 (2.8%) for Panama (Fig. 1).These values were greater than previously thought (La Marca et al. 2010).Lynch and Suárez-Mayorga ( 2001) mentioned that B. pugnax was a near-Colombian endemic species; however, our results suggest that this taxon is more extensively distributed in Venezuela.
Our SDM predicts the presence of B. pugnax in 14 ecosystems (Olson et al. 2001) for Venezuela; 4 of them encompass ~90% of its total area estimated in the country.The most extensive ecosystems were the savannah (~190,250 km 2 ; 48.25% of the extent areas), the dry forests (~84,500 km 2 ; 21.44%), the xeric shrublands forests (~63,000 km 2 ; 15.99%), and the Catatumbo moist forests (15,600 km 2 ; 3.96%).It is important to note that our SDM suggests the presence of B. pugnax in additional ecosystems (including moist forests, montane forests, mangroves, wetlands, and swamp) throughout Venezuelan states: Lara, Carabobo, Miranda, Sucre, and Bolívar.All these predictions will need a validation at field.However, the presence in these regions is expected on the base of habitat type and continuity reported in the areas, which is congruent with the species' biology (La Marca et al. 2010).
Additionally, the SDM suggest lowest suitability values throughout the Cordillera de Mérida, the Cordillera Oriental (Eastern Andes in Colombia), and the Sierra de Perijá (northern border between Venezuela and Colombia), which could be interpreted as potential geographic and ecological barriers for the dispersion of B. pugnax (Fig. 1).Hence, the change of physical variables along the elevational gradient in these mountains could constrain the spread of individuals through them (Janzen 1967).This is congruent with the allozymatic differences found between populations from northwest and southeast of Cordillera de Mérida (Nava 2005).However, the current evidence let us to treat all populations of B. pugnax as a single species.Nevertheless, these disjunct populations could merit the recognition as Evolutionarily Significant Units (ESUs) in the future, which will be important in a biology conservation context (Moritz 1994).
Finally, our results encourage the need to continue studying the biology of Boana pugnax, providing an ecological framework of where to focus the future survey efforts in Venezuela, as well as in Colombia and Panama.
Modeled distributions have the advantage of filling gaps that point-based distributions present as a result of the necessarily incomplete sampling (Peterson 2001, Mota-Vargas andRojas-Soto 2012), but at same time identify the most important variables for the species persistence (Soberón andPeterson 2005, Elith et al. 2011).Furthermore, this technique provides better results in terms of spatial and numerical sensitivity as well as lower values of omission and a moderate extent of predicted areas; therefore, are widely used in ecology, evolution, conservation, and management (e.g., Soberón and Peterson 2005, Mota-Vargas and Rojas-Soto 2012, Ortega-Andrade et al. 2015).Additionally, this study illustrates the importance of the museums specimens as a source of data (see Rocha et al. 2014) to increase the knowledge about species' geographic distribution.Appendix, Table A1.Continued.

Figure 2 .
Figure 2. Boana pugnax adult male in life (ULABG 7739) from Caño San Mateo (South of Maracaibo Lake Basin, Mérida state, Venezuela). A. Dorsolateral view.B. Ventral view showing the blackish coloration at throat level.C. Color pattern of the flank of body.D. Ventral view at thigh, showing its color pattern in preservative.Photos: Ivan Mendoza (A-C) and Moisés Escalona (D).
Soberón and Peterson 2005, for model calibration (or M sensu BAM diagram; seeSoberón and Peterson 2005,

Table 2 .
Summary of the selected, not-correlated, environmental variables with relative contributions (%) to the potential distribution model of Boana pugnax using MaxEnt 3.3.3k.